diff --git a/.gitignore b/.gitignore new file mode 100644 index 00000000..cc75c920 --- /dev/null +++ b/.gitignore @@ -0,0 +1,5 @@ +*.py[cod] +*.so +*.egg +*.egg-info +*.DS_Store diff --git a/Code/1_data_prepare/1_1_cifar10_to_png.py b/Code/1_data_prepare/1_1_cifar10_to_png.py index 9df14994..047944e5 100644 --- a/Code/1_data_prepare/1_1_cifar10_to_png.py +++ b/Code/1_data_prepare/1_1_cifar10_to_png.py @@ -3,22 +3,23 @@ 将cifar10的data_batch_12345 转换成 png格式的图片 每个类别单独存放在一个文件夹，文件夹名称为0-9 """ -from scipy.misc import imsave +from imageio import imwrite import numpy as np import os import pickle -data_dir = '../../Data/cifar-10-batches-py/' -train_o_dir = '../../Data/cifar-10-png/raw_train/' -test_o_dir = '../../Data/cifar-10-png/raw_test/' + +base_dir = "D:/python 11/新建文件夹/practise/pytorch" #修改为当前Data 目录所在的绝对路径 +data_dir = os.path.join(base_dir, "Data", "cifar-10-batches-py") +train_o_dir = os.path.join( base_dir, "Data", "cifar-10-png", "raw_train") +test_o_dir = os.path.join( base_dir, "Data", "cifar-10-png", "raw_test") Train = False # 不解压训练集，仅解压测试集 # 解压缩，返回解压后的字典 def unpickle(file): - fo = open(file, 'rb') - dict_ = pickle.load(fo, encoding='bytes') - fo.close() + with open(file, 'rb') as fo: + dict_ = pickle.load(fo, encoding='bytes') return dict_ def my_mkdir(my_dir): @@ -30,7 +31,7 @@ def my_mkdir(my_dir): if __name__ == '__main__': if Train: for j in range(1, 6): - data_path = data_dir + "data_batch_" + str(j) # data_batch_12345 + data_path = os.path.join(data_dir, "data_batch_" + str(j)) # data_batch_12345 train_data = unpickle(data_path) print(data_path + " is loading...") @@ -44,13 +45,13 @@ def my_mkdir(my_dir): img_name = label_num + '_' + str(i + (j - 1)*10000) + '.png' img_path = os.path.join(o_dir, img_name) - imsave(img_path, img) + imwrite(img_path, img) print(data_path + " loaded.") print("test_batch is loading...") # 生成测试集图片 - test_data_path = data_dir + "test_batch" + test_data_path = os.path.join(data_dir, "test_batch") test_data = unpickle(test_data_path) for i in range(0, 10000): img = np.reshape(test_data[b'data'][i], (3, 32, 32)) @@ -62,6 +63,6 @@ def my_mkdir(my_dir): img_name = label_num + '_' + str(i) + '.png' img_path = os.path.join(o_dir, img_name) - imsave(img_path, img) + imwrite(img_path, img) print("test_batch loaded.") diff --git a/Code/1_data_prepare/1_2_split_dataset.py b/Code/1_data_prepare/1_2_split_dataset.py index 21db34e7..1ca426bd 100644 --- a/Code/1_data_prepare/1_2_split_dataset.py +++ b/Code/1_data_prepare/1_2_split_dataset.py @@ -8,10 +8,10 @@ import random import shutil -dataset_dir = '../../Data/cifar-10-png/raw_test/' -train_dir = '../../Data/train/' -valid_dir = '../../Data/valid/' -test_dir = '../../Data/test/' +dataset_dir = os.path.join("..", "..", "Data", "cifar-10-png", "raw_test") +train_dir = os.path.join("..", "..", "Data", "train") +valid_dir = os.path.join("..", "..", "Data", "valid") +test_dir = os.path.join("..", "..", "Data", "test") train_per = 0.8 valid_per = 0.1 @@ -27,7 +27,7 @@ def makedir(new_dir): for root, dirs, files in os.walk(dataset_dir): for sDir in dirs: - imgs_list = glob.glob(os.path.join(root, sDir)+'/*.png') + imgs_list = glob.glob(os.path.join(root, sDir, '*.png')) random.seed(666) random.shuffle(imgs_list) imgs_num = len(imgs_list) @@ -37,14 +37,14 @@ def makedir(new_dir): for i in range(imgs_num): if i < train_point: - out_dir = train_dir + sDir + '/' + out_dir = os.path.join(train_dir, sDir) elif i < valid_point: - out_dir = valid_dir + sDir + '/' + out_dir = os.path.join(valid_dir, sDir) else: - out_dir = test_dir + sDir + '/' + out_dir = os.path.join(test_dir, sDir) makedir(out_dir) - out_path = out_dir + os.path.split(imgs_list[i])[-1] + out_path = os.path.join(out_dir, os.path.split(imgs_list[i])[-1]) shutil.copy(imgs_list[i], out_path) print('Class:{}, train:{}, valid:{}, test:{}'.format(sDir, train_point, valid_point-train_point, imgs_num-valid_point)) diff --git a/Code/1_data_prepare/1_3_generate_txt.py b/Code/1_data_prepare/1_3_generate_txt.py index c588b72e..057f0d6e 100644 --- a/Code/1_data_prepare/1_3_generate_txt.py +++ b/Code/1_data_prepare/1_3_generate_txt.py @@ -4,11 +4,11 @@ 为数据集生成对应的txt文件 ''' -train_txt_path = '../../Data/train.txt' -train_dir = '../../Data/train/' +train_txt_path = os.path.join("..", "..", "Data", "train.txt") +train_dir = os.path.join("..", "..", "Data", "train") -valid_txt_path = '../../Data/valid.txt' -valid_dir = '../../Data/valid/' +valid_txt_path = os.path.join("..", "..", "Data", "valid.txt") +valid_dir = os.path.join("..", "..", "Data", "valid") def gen_txt(txt_path, img_dir): @@ -30,4 +30,5 @@ def gen_txt(txt_path, img_dir): if __name__ == '__main__': gen_txt(train_txt_path, train_dir) - gen_txt(valid_txt_path, valid_dir) \ No newline at end of file + gen_txt(valid_txt_path, valid_dir) + diff --git a/Code/1_data_prepare/1_5_compute_mean.py b/Code/1_data_prepare/1_5_compute_mean.py index 28cb90e9..721a7d95 100644 --- a/Code/1_data_prepare/1_5_compute_mean.py +++ b/Code/1_data_prepare/1_5_compute_mean.py @@ -3,6 +3,7 @@ import numpy as np import cv2 import random +import os """ 随机挑选CNum张图片，进行按通道计算均值mean和标准差std @@ -10,7 +11,7 @@ """ -train_txt_path = '../../Data/train.txt' +train_txt_path = os.path.join("..", "..", "Data/train.txt") CNum = 2000 # 挑选多少图片进行计算 diff --git a/Code/2_model/2_finetune.py b/Code/2_model/2_finetune.py index 90407e8f..c6307c4c 100644 --- a/Code/2_model/2_finetune.py +++ b/Code/2_model/2_finetune.py @@ -14,8 +14,8 @@ from utils.utils import MyDataset, validate, show_confMat from datetime import datetime -train_txt_path = '../../Data/train.txt' -valid_txt_path = '../../Data/valid.txt' +train_txt_path = os.path.join("..", "..", "Data", "train.txt") +valid_txt_path = os.path.join("..", "..", "Data", "valid.txt") classes_name = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'] @@ -25,7 +25,7 @@ max_epoch = 1 # log -result_dir = '../../Result/' +result_dir = os.path.join("..", "..", "Result") now_time = datetime.now() time_str = datetime.strftime(now_time, '%m-%d_%H-%M-%S') diff --git a/Code/4_viewer/1_tensorboardX_demo.py b/Code/4_viewer/1_tensorboardX_demo.py index 9da8523b..acb60ba4 100644 --- a/Code/4_viewer/1_tensorboardX_demo.py +++ b/Code/4_viewer/1_tensorboardX_demo.py @@ -1,4 +1,5 @@ # coding: utf-8 +import os import torch import torchvision.utils as vutils import numpy as np @@ -7,7 +8,7 @@ from tensorboardX import SummaryWriter resnet18 = models.resnet18(False) -writer = SummaryWriter('../../Result/runs') +writer = SummaryWriter(os.path.join("..", "..", "Result", "runs")) sample_rate = 44100 freqs = [262, 294, 330, 349, 392, 440, 440, 440, 440, 440, 440] @@ -23,10 +24,10 @@ s1 = torch.rand(1) # value to keep s2 = torch.rand(1) # data grouping by `slash` - writer.add_scalar('data/scalar_systemtime', s1[0], n_iter) + writer.add_scalar(os.path.join("data", "scalar_systemtime"), s1[0], n_iter) # data grouping by `slash` - writer.add_scalar('data/scalar_customtime', s1[0], n_iter, walltime=n_iter) - writer.add_scalars('data/scalar_group', {"xsinx": n_iter * np.sin(n_iter), + writer.add_scalar(os.path.join("data", "scalar_customtime"), s1[0], n_iter, walltime=n_iter) + writer.add_scalars(os.path.join("data", "scalar_group"), {"xsinx": n_iter * np.sin(n_iter), "xcosx": n_iter * np.cos(n_iter), "arctanx": np.arctan(n_iter)}, n_iter) x = torch.rand(32, 3, 64, 64) # output from network @@ -56,15 +57,15 @@ precision, recall, n_iter) # export scalar data to JSON for external processing -writer.export_scalars_to_json("../../Result/all_scalars.json") +writer.export_scalars_to_json(os.path.join("..", "..", "Result", "all_scalars.json")) -dataset = datasets.MNIST('../../Data/mnist', train=False, download=True) +dataset = datasets.MNIST(os.path.join("..", "..", "Data", "mnist"), train=False, download=True) images = dataset.test_data[:100].float() label = dataset.test_labels[:100] features = images.view(100, 784) writer.add_embedding(features, metadata=label, label_img=images.unsqueeze(1)) writer.add_embedding(features, global_step=1, tag='noMetadata') -dataset = datasets.MNIST('../../Data/mnist', train=True, download=True) +dataset = datasets.MNIST(os.path.join("..", "..", "Data", "mnist"), train=True, download=True) images_train = dataset.train_data[:100].float() labels_train = dataset.train_labels[:100] features_train = images_train.view(100, 784) diff --git a/Code/4_viewer/2_visual_weights.py b/Code/4_viewer/2_visual_weights.py index cc1e340d..c11d9ea0 100644 --- a/Code/4_viewer/2_visual_weights.py +++ b/Code/4_viewer/2_visual_weights.py @@ -1,4 +1,5 @@ # coding: utf-8 +import os import torch import torchvision.utils as vutils from tensorboardX import SummaryWriter @@ -42,10 +43,10 @@ def initialize_weights(self): net = Net() # 创建一个网络 -pretrained_dict = torch.load('../2_model/net_params.pkl') +pretrained_dict = torch.load(os.path.join("..", "2_model", "net_params.pkl")) net.load_state_dict(pretrained_dict) -writer = SummaryWriter(log_dir='../../Result/visual_weights') +writer = SummaryWriter(log_dir=os.path.join("..", "..", "Result", "visual_weights")) params = net.state_dict() for k, v in params.items(): if 'conv' in k and 'weight' in k: diff --git a/Code/4_viewer/3_visual_featuremaps.py b/Code/4_viewer/3_visual_featuremaps.py index 07ef0aaf..23b175e9 100644 --- a/Code/4_viewer/3_visual_featuremaps.py +++ b/Code/4_viewer/3_visual_featuremaps.py @@ -1,4 +1,5 @@ # coding: utf-8 +import os import torch import torchvision.utils as vutils import numpy as np @@ -12,9 +13,9 @@ vis_layer = 'conv1' -log_dir = '../../Result/visual_featuremaps' -txt_path = '../../Data/visual.txt' -pretrained_path = '../../Data/net_params_72p.pkl' +log_dir = os.path.join("..", "..", "Result", "visual_featuremaps") +txt_path = os.path.join("..", "..", "Data", "visual.txt") +pretrained_path = os.path.join("..", "..", "Data", "net_params_72p.pkl") net = Net() pretrained_dict = torch.load(pretrained_path) diff --git a/Code/4_viewer/4_hist_grad_weight.py b/Code/4_viewer/4_hist_grad_weight.py index 677df784..79b00d8e 100644 --- a/Code/4_viewer/4_hist_grad_weight.py +++ b/Code/4_viewer/4_hist_grad_weight.py @@ -9,13 +9,14 @@ import torch.nn as nn import torch.optim as optim import sys +import os sys.path.append("..") from utils.utils import MyDataset, validate, show_confMat, Net from tensorboardX import SummaryWriter from datetime import datetime -train_txt_path = '../../Data/train.txt' -valid_txt_path = '../../Data/valid.txt' +train_txt_path = os.path.join("..", "..", "Data", "train.txt") +valid_txt_path = os.path.join("..", "..", "Data", "valid.txt") classes_name = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'] @@ -25,7 +26,7 @@ max_epoch = 1 # log -log_dir = '../../Result/hist_grad_weight' +log_dir = os.path.join("..", "..", "Result", "hist_grad_weight") writer = SummaryWriter(log_dir=log_dir) diff --git a/Code/4_viewer/5_Show_ConfMat.py b/Code/4_viewer/5_Show_ConfMat.py index f86f2a23..a3d23b52 100644 --- a/Code/4_viewer/5_Show_ConfMat.py +++ b/Code/4_viewer/5_Show_ConfMat.py @@ -42,5 +42,5 @@ def show_confMat(confusion_mat, classes_name, set_name, out_dir): if __name__ == '__main__': - print('QQ group: {}, password: {}'.format(671103375, 2018)) + print('QQ group: {} or {} or {} or {}, password: {}'.format(671103375, 773031536, 514974779, 854620826, 2018)) diff --git a/Code/4_viewer/6_hook_for_grad_cam.py b/Code/4_viewer/6_hook_for_grad_cam.py new file mode 100644 index 00000000..faf4b880 --- /dev/null +++ b/Code/4_viewer/6_hook_for_grad_cam.py @@ -0,0 +1,184 @@ +# coding: utf-8 +""" +通过实现Grad-CAM学习module中的forward_hook和backward_hook函数 +""" +import cv2 +import os +import numpy as np +from PIL import Image +import torch +import torch.nn as nn +import torch.nn.functional as F +import torchvision.transforms as transforms + + +class Net(nn.Module): + def __init__(self): + super(Net, self).__init__() + self.conv1 = nn.Conv2d(3, 6, 5) + self.pool1 = nn.MaxPool2d(2, 2) + self.conv2 = nn.Conv2d(6, 16, 5) + self.pool2 = nn.MaxPool2d(2, 2) + self.fc1 = nn.Linear(16 * 5 * 5, 120) + self.fc2 = nn.Linear(120, 84) + self.fc3 = nn.Linear(84, 10) + + def forward(self, x): + x = self.pool1(F.relu(self.conv1(x))) + x = self.pool1(F.relu(self.conv2(x))) + x = x.view(-1, 16 * 5 * 5) + x = F.relu(self.fc1(x)) + x = F.relu(self.fc2(x)) + x = self.fc3(x) + return x + + +def img_transform(img_in, transform): + """ + 将img进行预处理，并转换成模型输入所需的形式—— B*C*H*W + :param img_roi: np.array + :return: + """ + img = img_in.copy() + img = Image.fromarray(np.uint8(img)) + img = transform(img) + img = img.unsqueeze(0) # C*H*W --> B*C*H*W + return img + + +def img_preprocess(img_in): + """ + 读取图片，转为模型可读的形式 + :param img_in: ndarray, [H, W, C] + :return: PIL.image + """ + img = img_in.copy() + img = cv2.resize(img,(32, 32)) + img = img[:, :, ::-1] # BGR --> RGB + transform = transforms.Compose([ + transforms.ToTensor(), + transforms.Normalize([0.4948052, 0.48568845, 0.44682974], [0.24580306, 0.24236229, 0.2603115]) + ]) + img_input = img_transform(img, transform) + return img_input + + +def backward_hook(module, grad_in, grad_out): + grad_block.append(grad_out[0].detach()) + + +def farward_hook(module, input, output): + fmap_block.append(output) + + +def show_cam_on_image(img, mask, out_dir): + heatmap = cv2.applyColorMap(np.uint8(255*mask), cv2.COLORMAP_JET) + heatmap = np.float32(heatmap) / 255 + cam = heatmap + np.float32(img) + cam = cam / np.max(cam) + + path_cam_img = os.path.join(out_dir, "cam.jpg") + path_raw_img = os.path.join(out_dir, "raw.jpg") + if not os.path.exists(out_dir): + os.makedirs(out_dir) + cv2.imwrite(path_cam_img, np.uint8(255 * cam)) + cv2.imwrite(path_raw_img, np.uint8(255 * img)) + + +def comp_class_vec(ouput_vec, index=None): + """ + 计算类向量 + :param ouput_vec: tensor + :param index: int，指定类别 + :return: tensor + """ + if not index: + index = np.argmax(ouput_vec.cpu().data.numpy()) + else: + index = np.array(index) + index = index[np.newaxis, np.newaxis] + index = torch.from_numpy(index) + one_hot = torch.zeros(1, 10).scatter_(1, index, 1) + one_hot.requires_grad = True + class_vec = torch.sum(one_hot * output) # one_hot = 11.8605 + + return class_vec + + +def gen_cam(feature_map, grads): + """ + 依据梯度和特征图，生成cam + :param feature_map: np.array， in [C, H, W] + :param grads: np.array， in [C, H, W] + :return: np.array, [H, W] + """ + cam = np.zeros(feature_map.shape[1:], dtype=np.float32) # cam shape (H, W) + + weights = np.mean(grads, axis=(1, 2)) # + + for i, w in enumerate(weights): + cam += w * feature_map[i, :, :] + + cam = np.maximum(cam, 0) + cam = cv2.resize(cam, (32, 32)) + cam -= np.min(cam) + cam /= np.max(cam) + + return cam + + +if __name__ == '__main__': + + BASE_DIR = os.path.dirname(os.path.abspath(__file__)) + path_img = os.path.join(BASE_DIR, "..", "..", "Data", "cam_img", "test_img_8.png") + path_net = os.path.join(BASE_DIR, "..", "..", "Data", "net_params_72p.pkl") + output_dir = os.path.join(BASE_DIR, "..", "..", "Result", "backward_hook_cam") + + classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck') + fmap_block = list() + grad_block = list() + + # 图片读取；网络加载 + img = cv2.imread(path_img, 1) # H*W*C + img_input = img_preprocess(img) + net = Net() + net.load_state_dict(torch.load(path_net)) + + # 注册hook + net.conv2.register_forward_hook(farward_hook) + net.conv2.register_backward_hook(backward_hook) + + # forward + output = net(img_input) + idx = np.argmax(output.cpu().data.numpy()) + print("predict: {}".format(classes[idx])) + + # backward + net.zero_grad() + class_loss = comp_class_vec(output) + class_loss.backward() + + # 生成cam + grads_val = grad_block[0].cpu().data.numpy().squeeze() + fmap = fmap_block[0].cpu().data.numpy().squeeze() + cam = gen_cam(fmap, grads_val) + + # 保存cam图片 + img_show = np.float32(cv2.resize(img, (32, 32))) / 255 + show_cam_on_image(img_show, cam, output_dir) + + + + + + + + + + + + + + + + diff --git a/Code/main_training/main.py b/Code/main_training/main.py index 13f3cf72..8fd0c526 100644 --- a/Code/main_training/main.py +++ b/Code/main_training/main.py @@ -15,8 +15,8 @@ from tensorboardX import SummaryWriter from datetime import datetime -train_txt_path = '../../Data/train.txt' -valid_txt_path = '../../Data/valid.txt' +train_txt_path = os.path.join("..", "..", "Data", "train.txt") +valid_txt_path = os.path.join("..", "..", "Data", "valid.txt") classes_name = ['plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'] @@ -26,7 +26,7 @@ max_epoch = 1 # log -result_dir = '../../Result/' +result_dir = os.path.join("..", "..", "Result") now_time = datetime.now() time_str = datetime.strftime(now_time, '%m-%d_%H-%M-%S') diff --git a/Data/alipay.jpg b/Data/alipay.jpg new file mode 100644 index 00000000..4ecdf146 Binary files /dev/null and b/Data/alipay.jpg differ diff --git a/Data/cam_img/test_img_1.png b/Data/cam_img/test_img_1.png new file mode 100644 index 00000000..bbdb6579 Binary files /dev/null and b/Data/cam_img/test_img_1.png differ diff --git a/Data/cam_img/test_img_2.png b/Data/cam_img/test_img_2.png new file mode 100644 index 00000000..8da5599e Binary files /dev/null and b/Data/cam_img/test_img_2.png differ diff --git a/Data/cam_img/test_img_3.png b/Data/cam_img/test_img_3.png new file mode 100644 index 00000000..bb054813 Binary files /dev/null and b/Data/cam_img/test_img_3.png differ diff --git a/Data/cam_img/test_img_4.png b/Data/cam_img/test_img_4.png new file mode 100644 index 00000000..a3762d04 Binary files /dev/null and b/Data/cam_img/test_img_4.png differ diff --git a/Data/cam_img/test_img_5.png b/Data/cam_img/test_img_5.png new file mode 100644 index 00000000..1af7f3ec Binary files /dev/null and b/Data/cam_img/test_img_5.png differ diff --git a/Data/cam_img/test_img_6.png b/Data/cam_img/test_img_6.png new file mode 100644 index 00000000..9c6e7f22 Binary files /dev/null and b/Data/cam_img/test_img_6.png differ diff --git a/Data/cam_img/test_img_7.png b/Data/cam_img/test_img_7.png new file mode 100644 index 00000000..5a50b63e Binary files /dev/null and b/Data/cam_img/test_img_7.png differ diff --git a/Data/cam_img/test_img_8.png b/Data/cam_img/test_img_8.png new file mode 100644 index 00000000..776e12e9 Binary files /dev/null and b/Data/cam_img/test_img_8.png differ diff --git a/Data/wechat.jpg b/Data/wechat.jpg new file mode 100644 index 00000000..79e43103 Binary files /dev/null and b/Data/wechat.jpg differ diff --git a/readme.md b/readme.md index b16b1fbd..50db95df 100644 --- a/readme.md +++ b/readme.md @@ -1,11 +1,23 @@ # Pytorch模型训练实用教程 -![Image text](https://github.com/tensor-yu/PyTorch_Tutorial/blob/master/Data/cover.png) +Image text + +--- + +📢：《PyTorch实用教程》（第二版）已开源，欢迎阅读：https://tingsongyu.github.io/PyTorch-Tutorial-2nd/ + +📢：《PyTorch实用教程》（第二版）已开源，欢迎阅读：https://tingsongyu.github.io/PyTorch-Tutorial-2nd/ + +📢：《PyTorch实用教程》（第二版）已开源，欢迎阅读：https://tingsongyu.github.io/PyTorch-Tutorial-2nd/ + +第二版新增丰富的**深度学习应用案例**和**推理部署框架**，包括CV、NLP和LLM的十多个实战项目，以及ONNX和TensorRT的教程。 # 1.简介 + 本代码为教程——《Pytorch模型训练实用教程》中配套代码；
《Pytorch模型训练实用教程》可通过如下方式获取：
+ 1. https://github.com/tensor-yu/PyTorch_Tutorial/tree/master/Data
-2. QQ群：671103375
+2. QQ群： 四群：854620826
# 2.环境配置 @@ -24,7 +36,17 @@ https://pytorch.org/get-started/locally/ # 3.问题反馈 若发现任何问题和改进意见，请您随时联系我。
联系方式：yts3221@126.com
-读者qq群：671103375 +读者qq群： + +​ 一群：671103375 (已满)
+ +​ 二群：773031536 (已满）
+ +​ 三群：514974779 (已满）
+ +​ 四群：854620826(已满) + +​ 五群：1021300804 # 4.修改记录 0.0.5： @@ -32,3 +54,20 @@ https://pytorch.org/get-started/locally/ 2. 2.3小节删除注释； 3. 修改权值初始化杂谈中的理解错误； 4. 全文代码缩进。 + +--- + +如果本教程对你有帮助😀😀，请作者喝杯茶吧🍵🍵🥂🥂 + +WeChat： Alipay: + + + + + +--- + + +## Stargazers over time + +[![Stargazers over time](https://starchart.cc/TingsongYu/PyTorch_Tutorial.svg)](https://starchart.cc/TingsongYu/PyTorch_Tutorial)